GraphChi: How a Mac Mini outperformed a 1,636-node Hadoop cluster


This is a pretty interesting benchmark study, although the headline is a bit misleading because Hadoop isn’t really optimized for graph analysis. When you look at comparisons to Spark, GraphLab and other platforms, it seems the decision of what to choose might come down to data volume, acceptable latency and cost, especially when considered against the value of that graph workload. Projects like Giraph and other YARN-enabled engines might make Hadoop look better, too.

Comments are closed.